Re: [PATCH] Use realparamtab and don't dereference special parameters in zle code

[Philippe Altherr <philippe.altherr@xxxxxxxxx>]

Normally you would explain a change like this with motivations of some
sort. Why is this correct? When is paramtab possibly some other
parameter table than realparamtab? Is this change making things less
flexible? Are you solving an actual problem? These are things you've
presumably already investigated, so you should tell us about it so we
don't have to duplicate the work.

The patch does more of the same as workers/54475, which was already committed. That's why I didn't elaborate.

The patch does two things:

• Replace some calls to getnode() with calls to getnode2(). The main difference between the two is that the latter returns the parameter found in the table as is, while the former dereferences any named reference it finds. In some places, you only need the latter. Sometimes the former is not just unnecessary but problematic, see workers/54475 for an example.

• Replace paramtab with realparamtab in places that intend to refer to the parameter table. More on this below.

The two changes are in principle independent of each other but because they often touch the same lines of code it's easier to do it at the same time. If I need to understand a piece of code for one then it's not much more work to also understand it for the other. Doing the changes in the same patch avoids merge conflicts.

realparamtab vs paramtab

The two global variables are defined in params.c. Originally there was only paramtab, which contained the parameter table. When associative tables were added to Zsh, it turned out that many functions that worked on the parameter table could also be useful for associative tables. In order to reuse them instead of duplicating them, the global variable realparamtab was added to contain the parameter table and the global paramtab was repurposed to either contain the parameter table or an associative table. Thus, one can reuse a function originally designed for the parameter table to work on an associative table by initializing paramtab with that associative table before calling the function and then restoring paramtab to its previous value after the call. At startup, paramtab is initialized with realparamtab.

This (imo horrible) hack had the great advantage that it enabled the reuse of many functions with very little code changes. An unfortunate consequence is that it instantly turned a lot of code into code that seemingly works both for the parameter table and for associative tables, while in reality it only makes sense for the former. Replacing paramtab with realparamtab makes that clear.

The vast majority of functions that actually make sense for both usages are in params.c. Ultimately paramtab should only be used in these functions. My secret hope is to even get rid of the paramtab global variable by adding an extra parameter to these functions but that will depend on how many such functions there are. If it's just a few, it would be a net win but if there are many, it may not be worth it.

Philippe